Middlebury College Undergraduate Students Reproduce Three Spatial Analysis Studies in Health and Hazards

Keywords: Reproducibility, GIScience, Climate Vulnerability, COVID-19
Abstract Type: Paper Abstract

Abstract

Middlebury College geography students reproduced three health/hazards spatial analysis studies using open source R or Python code. The studies included 1) climate vulnerability model of Malawi, 2) spatial association of COVID-19 incidence and people with disabilities in U.S. counties, and 3) accessibility of COVID-19 health care resources in Chicago. In our reproduction studies, we attempted to generate identical results as a prior study using identical data and methods, with the goal of assessing the prior study's credibility and internal validity. We acquired data and wrote new code to complete reproductions for studies with missing components. We also adjusted and reanalyzed study designs to assess error and uncertainty and to improve conceptualizations and methods. We published reproducible research compendia for all three studies on GitHub and OSF, thereby increasing the reproducibility of all three studies.

We reproduced inexact but similar results for all three studies, and reanalyzed the studies to improve conceptualizations, methods, and data visualizations. Our reproductions highlighted the limitations of traditional narrative research articles for communicating complex spatial analysis methodologies. We also highlighted recurrent challenges in spatial analysis research, including data errors, the modifiable aerial unit problem, boundary effects, ecological fallacies with aggregated data, uncertainties associated with integrating different data sources, and construct validity of indicator data. Most importantly, we demonstrate how geography students can learn reproducible research methods and contribute to the geographic sciences by completing reproductions of recently published studies.